Load 3D Model from Asset

ComfyTV isn't just a video suite - the author's roots are in 3D (they shipped ComfyUI-InstantMesh before this), and the pack carries a Scene3D stage with multi-camera viewports, PBR materials, and mesh tools. Every one of those flows starts with a model on a wire. Load 3D Model from Asset is that start: pick a model from your project library and it comes out on a COMFYTV_MODEL output, with a preview image to go with it.

The nice touch is the second output. The node body embeds a real 3D preview viewport, and whatever angle you leave it at becomes a COMFYTV_IMAGE snapshot. So one stage hands you both the model and a "hero shot" of it - handy when you want to drop a rendered thumbnail into a contact sheet or drive an image stage off the model's appearance without rendering the scene first.

How it works

Selecting a model in the node body writes its payload URL into the internal asset_url input, and the stage emits it as the model output (type COMFYTV_MODEL). Meanwhile the preview viewport's current view is captured to captured_image - an internal /view? URL, per its tooltip - which becomes the image output. The project_id, parent_output_id, and asset_id inputs are frontend bookkeeping for lineage and debugging; leave them alone. category just remembers your last library filter.

Unlike the image/audio/video asset loaders, this one has two real outputs to think about:

• model (COMFYTV_MODEL) - wire this into Scene3D, the geometry workshop, or the PBR material stage.
• image (COMFYTV_IMAGE) - the preview snapshot, for anything that wants a picture of the model.

Installing ComfyTV

It's one pack with ~190 stages, so installing this node is installing ComfyTV:

cd ComfyUI/custom_nodes
git clone https://github.com/jtydhr88/ComfyTV

Restart the ComfyUI backend fully (not just refresh the browser) and stages appear under the ComfyTV category; ComfyUI Manager finds it by searching "ComfyTV". It ships zero extra Python dependencies - it uses what ComfyUI already bundles, and the 3D preview runs in the browser. No model downloads for this loader, though the 3D generation workflows in the pack do reference model files (see the pack's models.md for those).

The README's install warning applies double here because of that Preview-only-in-the-right-instance behavior: on macOS, ComfyUI Desktop, or multi-install setups, the relative cd can clone into the wrong ComfyUI. Read the running instance's path from the startup log, clone into that instance's custom_nodes by absolute path (quotes if it has spaces/parentheses), verify ComfyTV/__init__.py is top-level, and restart the backend - not just the tab.

Common issues

• Preview renders black or the stage errors. The 3D viewport needs a model actually selected. If the library is empty, you're likely in the wrong project - assets are per-project.
• COMFYTV_MODEL won't connect to another pack's model nodes. Expected; the typed wire is pack-internal. Bridge nodes exist for plugging third-party plugins into a ComfyTV pipeline.
• The image output is a stale angle. It's whatever the viewport last showed. Spin the model the way you want the thumbnail before you run.

If your ComfyTV work is image and video only, you can skip this stage entirely. The day you want a 3D shot in a workflow, though, this is the cleanest on-ramp in the pack.